common-lisp-jupyter icon indicating copy to clipboard operation
common-lisp-jupyter copied to clipboard

zmq.h not found

Open hsmyers opened this issue 1 year ago • 7 comments

After executing "ros install common-lisp-jupyter," I was presented with a host of error text that boiled down to "zmq.h not found." This running under Windows 11 (bleeding edge) with a fresh install of roswell via scoop. I was not aware that roswell would install a duplicate version of sbcl, but don't see that as a problem. I saw that some time back, you were (I think) able to solve this for a Mac installation, but I did not see anything that seemed a good fit. Help?

hsmyers avatar Jan 23 '24 19:01 hsmyers

You need to have ZeroMQ installed per the installation requirements: https://yitzchak.github.io/common-lisp-jupyter/install#requirements

I don't use Windows regularly so I am not an expert in this, but there is an example of this being done in the GitHub workflow via Conda. There are other ways to do this if you are using a different package manager.

https://github.com/yitzchak/common-lisp-jupyter/blob/63280ccda8e334f5dc79823662f0a3ded42ffac3/.github/workflows/ci.yml#L34-L39

yitzchak avatar Jan 23 '24 20:01 yitzchak

I've no experience using conda. I suppose that's because I don't use Python. You mentioned others? I have scoop and chocolatey (choco), and whatever comes from windows, wget I think

hsmyers avatar Jan 23 '24 21:01 hsmyers

Windows user here. It works, but it's painful (as mentioned, primary/only author doesn't use MS Windows). Having recently setup everything for a CL notebook, again, the working formula, for me, is to keep everything in the MSYS2 environment. Python, ZeroMQ, etc. The only exception is conda -- I used the MS Windows install and then put it in my bash shell path along with with this snippet in .bashrc:

# This should really be avoided, as it brings in it's own gcc compiler toolchain.
# See: https://stackoverflow.com/questions/65572789/i-have-installed-msys2-and-anaconda3-then-how-can-i-use-conda-on-msys2
# >>> conda initialize >>>
# !! Contents within this block are managed by 'conda init' !!
eval "$('/c/Users/Nunez/miniconda3/Scripts/conda.exe' 'shell.bash' 'hook')"

Yes, it probably isn't the most efficient, but seems to be stable. Using a pure-Windows install was, for me, fragile and seemed to break on every CLJ release.

Oh, and the jupyter-lab executable seems to barf on not being able to find zmq.dll. I solve this by placing the executable in the MSYS2 bin directory.

It's probably possible to get all this fixed up 'properly', but I ran out of patience. Let us know how you go.

snunez1 avatar Jan 24 '24 00:01 snunez1

Thanks @snunez1

Unfortunately, ZeroMQ is required by the Jupyter protocol. If we had a CL-native implementation of ZeroMQ than we could forgo the CFFI dependency. I have looked into the specs for the protocol and it certainly seems possible using usocket. It would just take time to write from scratch. It would obviously benefit Windows the most. The benefit to Linux and Mac would be less, but it would mean that I could ditch static-vectors, which I really dislike.

yitzchak avatar Jan 24 '24 00:01 yitzchak

I have a very rough proof-of-concept that ditches the CFFI pzmq for a native Lisp ZeroMQ dependency. Much more work to be done before I'll even be able to make a decision whether to go in this direction.

yitzchak avatar Feb 04 '24 20:02 yitzchak

As a frustrated not-yet-user, let me encourage you!

Hugh S. Myers

On Sun, Feb 4, 2024 at 1:55 PM Tarn W. Burton @.***> wrote:

I have a very rough proof-of-concept that ditches the CFFI pzmq for a native Lisp ZeroMQ dependency. Much more work to be done before I'll even be able to make a decision whether to go in this direction.

— Reply to this email directly, view it on GitHub https://github.com/yitzchak/common-lisp-jupyter/issues/119#issuecomment-1925911409, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAHE7PRNJKMSBNYMOGYEM63YR7YUBAVCNFSM6AAAAABCHSYZAWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMRVHEYTCNBQHE . You are receiving this because you authored the thread.Message ID: @.***>

hsmyers avatar Feb 04 '24 21:02 hsmyers

I am still a long way from deciding whether or not to continue in this direction. I haven't tested the native CL ZeroMQ client with the two primary users of common-lisp-jupyter yet (Maxima and CANDO). For those that want to try out my current attempt you will need clones of the following in an ASDF findable location (configure the ASDF registry or put them in ~/quicklisp/local-projects).

  1. https://github.com/yitzchak/nontrivial-gray-streams
  2. https://github.com/yitzchak/nilmq

You'll also need the nilmq branch of common-lisp-jupyter.

yitzchak avatar Feb 10 '24 16:02 yitzchak